Semantic Web-Based Information Systems: State-of-the-Art ...

Bu ld ng Ontology Translat on Systems
From a human user’s point of view, the first WebODE definition is more legible; at
first glance, the user can see that the relation usesTransportMean is defined between
the concepts flight and airTransportMean, and that its maximum cardinality is one.
In the second case, the user must find and interpret the four components (the ad hoc
relation definition and the three formal axioms) to reach the same conclusion.
A similar conclusion can be obtained from an application point of view. Let us suppose
that we want to populate the ontology with an annotation tool. The behavior
of the annotation tool is different for both definitions. With the first definition, the
annotation tool will easily understand that its user interface cannot give users the
possibility of adding more than one instance of the relation, and that the drop-down
lists used for selecting the domain and range of a relation instance will show only
direct or indirect instances of the concepts flight and airTransportMean, respectively.
With the second definition, the annotation tool will allow creating more than one
relation instance from the same instance and will display all the ontology instances
in the drop-down lists instead of just presenting instances of flight and airTransport-
Mean, respectively. After that, the annotation tool will have to run the consistency
checker to detect inconsistencies in the ontology.
Relationships.Between.Ontology.Translation.Layers
Figure 5 shows an example of a transformation from the ontology platform WebODE
to the language OWL DL. In this example, we have to transform two ad hoc
relations with the same name (usesTransportMean) and with different domains and
ranges (a flight uses an airTransportMean, and a cityBus uses a bus). In OWL DL,
the scope of an object property is global to the ontology, and thus we cannot define
two different object properties with the same name. In this example, we show that
translation decisions have to be taken at all layers, and we also show how the decision
taken at one layer can influence the decisions to be made at the others, hence
showing the complexity of this task.
Option 1 is driven by semantics; to preserve semantics in the transformation, two
different object properties with different identifiers are defined. Option 2 is driven
by pragmatics; only one object property is defined from both ad hoc relations, since
we assume that they refer to the same meaning, but some knowledge is lost in the
transformation (the one related to the object property domain and range). Finally,
Option 3 also is driven by pragmatics, with more care on the semantics; again, only
one object property is defined, and its domain and range is more restricted than in
Option 2, although we still lose the exact correspondence between each domain
and range.
Copyright © 2007, Idea Group Inc. Copying or distributing in print or electronic forms without written permission
of Idea Group Inc. is prohibited.